My AI workflow for building a custom blog
The real challenge started when I wanted a comment section. Most modern plugins are a nightmare—they either force users to create developer accounts or spam them with tracking ads. I wanted a simple, old-school forum vibe: pick a nickname, enter a password, and post. Since I couldn't find a service that did this simply, I decided to build it from scratch.
The deployment process
I managed to build a database, an edge function for handling comments, spam protection, and email notifications in about a day. I didn't write the lines of code myself; I used a continuous feedback loop with AI:
1. Describe the exact behavior I wanted in plain English.
2. Let the AI generate the code.
3. Test the feature immediately.
4. Feed the errors back into the AI and repeat until it worked.
Real-world friction
It wasn't a perfect linear path. I hit a few walls that any beginner would:
- Wrong Tooling: I initially deployed a ready-made system, only to realize it forced logins. I ripped it out within an hour.
- Caching Issues: I panicked twice when updates didn't show up, only to realize it was just a cache issue. I had to learn to stop assuming I'd broken the entire site.
- Environmental Bugs: I thought my emoji reactions were broken, but it turned out my terminal was just mangling the characters. The bug is almost always simpler (and dumber) than you think.
This whole experience proved that you don't need to be a developer to ship functional software. You just need to be able to describe your requirements clearly, refuse the first "no" you encounter, and test your product like a skeptical customer. It turns a complex deployment into a series of manageable conversations with an LLM agent.